triform/cargoless
Template
1
0
Fork 0
Code Issues 1 Pull requests 18 Projects Releases Packages Wiki Activity Actions

feat(lanetree): classify no-author conflicts as resolvable rather than fatal #165

Open
triform-admin wants to merge 2 commits from agent/lane-generated-tree-resolver into main
pull from: agent/lane-generated-tree-resolver
merge into: triform:main
Conversation 0 Commits 2 Files changed 2 +330
triform-admin commented 2026-08-20 22:09:03 +00:00
Owner
Copy link

What

Adds the decision half of a lane conflict resolver: which conflicting paths the
lane may resolve automatically, and which it must eject. Pure logic + tests; it
is not yet wired into materialize, so this changes no lane behaviour.

Why

Ejections were being recovered by a janitor loop, not prevented. Some are not
anybody's conflict at all:

  • Racing image pins (the big one). A bot rewrites image tags on trunk
    continuously — 105 auto-bake commits on dev in three days (~35/day) —
    while enrolled PRs carry their own bake of the same manifests. Two bakes on
    the same newTag/digest lines conflict by construction. Mean PR survival
    (~37 min) is roughly the gap between bakes.
  • Committed codegen output, which is a pure function of sources that merged
    cleanly. tf-multiverse already resolves this class on promotion —
    scripts/promote-parity calls generated trees "the only auto-resolvable
    conflict class"     — but the lane never did.

Measured, not assumed

git merge-tree against live dev over 80 open PRs; 27 carry a true conflict.

rule conflicting PRs resolvable
codegen only 3 (11%)
+ racing-pin 14 (51%)

Ground truth on the four admin-app.yaml conflicts: each is a single
image: line — one dev-side digest vs four different PR-side digests, zero
other content. And 8 consecutive bake commits changed zero lines outside
newTag/digest/image/artifact-URL.

An earlier estimate of 85% was wrong — it counted changed files rather
than conflicting ones. The real leaders are hand-written source (49%) and
deployment pins (26%); codegen is 14%. The corrected numbers are recorded in
the module header so the next reader doesn't inherit the mistake.

Fail closed

  • Every path in a conflict set must qualify; one unknown path ejects the whole
    member, exactly as today.
  • portal/src/generated/ is excluded despite its name — it has no
    generator and is a hand-maintained fork (337 files); pinning it would discard
    human edits.
  • CAPABILITIES.md / circles meta have checkers but no confirmed regenerator,
    so they eject.
  • The rev-stamp filters check-codegen-drift.sh uses are carried here too;
    without them a byte-identity assertion reports false drift on every
    regenerated PR — ejecting exactly the population this saves.

Tests

14 new (662 total, 0 failures), including the known-negatives that matter:
hand-written source still ejects, a hand-edited replica count in a bake-written
manifest still ejects, an empty/unreadable hunk still ejects.

Not in this PR

Wiring into LaneTree::materialize at the existing conflict site
(lanetree.rs:162-182, where unmerged_paths is already read before
merge --abort), and the regeneration step itself.

Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com

## What Adds the decision half of a lane conflict resolver: which conflicting paths the lane may resolve automatically, and which it must eject. Pure logic + tests; it is **not yet wired into `materialize`**, so this changes no lane behaviour. ## Why Ejections were being recovered by a janitor loop, not prevented. Some are not anybody's conflict at all: - **Racing image pins (the big one).** A bot rewrites image tags on trunk continuously — **105 `auto-bake` commits on `dev` in three days (~35/day)** — while enrolled PRs carry their own bake of the same manifests. Two bakes on the same `newTag`/`digest` lines conflict by construction. Mean PR survival (~37 min) is roughly the gap between bakes. - **Committed codegen output**, which is a pure function of sources that merged cleanly. tf-multiverse already resolves this class on promotion — `scripts/promote-parity` calls generated trees *"the only auto-resolvable conflict class"* — but the lane never did. ## Measured, not assumed `git merge-tree` against live `dev` over 80 open PRs; 27 carry a true conflict. | rule | conflicting PRs resolvable | | --- | --- | | codegen only | 3 (11%) | | + racing-pin | **14 (51%)** | Ground truth on the four `admin-app.yaml` conflicts: each is a **single `image:` line** — one dev-side digest vs four different PR-side digests, zero other content. And 8 consecutive bake commits changed **zero** lines outside `newTag`/`digest`/`image`/artifact-URL. **An earlier estimate of 85% was wrong** — it counted *changed* files rather than *conflicting* ones. The real leaders are hand-written source (49%) and deployment pins (26%); codegen is 14%. The corrected numbers are recorded in the module header so the next reader doesn't inherit the mistake. ## Fail closed - Every path in a conflict set must qualify; one unknown path ejects the whole member, exactly as today. - `portal/src/generated/` is **excluded despite its name** — it has no generator and is a hand-maintained fork (337 files); pinning it would discard human edits. - `CAPABILITIES.md` / circles meta have checkers but no confirmed regenerator, so they eject. - The rev-stamp filters `check-codegen-drift.sh` uses are carried here too; without them a byte-identity assertion reports false drift on every regenerated PR — ejecting exactly the population this saves. ## Tests 14 new (662 total, 0 failures), including the known-negatives that matter: hand-written source still ejects, a hand-edited replica count in a bake-written manifest still ejects, an empty/unreadable hunk still ejects. ## Not in this PR Wiring into `LaneTree::materialize` at the existing conflict site (`lanetree.rs:162-182`, where `unmerged_paths` is already read before `merge --abort`), and the regeneration step itself. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> 
Some lane ejections collide only in committed codegen output — files that
are a pure function of sources that merged cleanly. tf-multiverse already
resolves this class on promotion (scripts/promote-parity calls generated
trees "the only auto-resolvable conflict class"); the lane never did.

This adds the decision half: a fail-closed allowlist of paths PROVEN
reproducible by scripts/ci/check-codegen-drift.sh, plus the two rev-stamp
filters that guard applies — without them a byte-identity assertion reports
false drift on every regenerated PR, i.e. ejects exactly the population this
saves.

Scope is deliberately narrow and measured, not assumed. Over 80 open PRs,
27 carry a true conflict; 8 of their 57 conflicting paths (14%) are
regenerable and 3 PRs (11%) conflict only in such files. An earlier estimate
of 85% counted changed files instead of conflicting ones and was wrong — the
real leaders are hand-written source (49%) and deployment pins (26%).

portal/src/generated/ is excluded on purpose: it has NO generator and is a
hand-maintained fork, so pinning it would discard human edits. CAPABILITIES.md
and circles meta have checkers but no confirmed regenerator, so they eject.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
feat(lanetree): resolve racing image-pin conflicts newer-wins
All checks were successful
ci / rust-analyzer latency harness (S1 / AC#2) (pull_request) Has been skipped
Details
ci / fmt (pull_request) Successful in 1m17s
Details
ci / build (pull_request) Successful in 2m57s
Details
ci / test (pull_request) Successful in 3m39s
Details
ci / clippy (pull_request) Successful in 3m49s
Details
ci / lane policy mutation proof (pull_request) Successful in 7m20s
Details
2c648ce0f6 
The bigger no-author conflict class is not codegen — it is deployment pins.
A bot rewrites image tags on trunk continuously (105 auto-bake commits on dev
in three days, ~35/day) while enrolled PRs carry their own bake of the same
manifests. Two bakes on the same newTag/digest lines conflict by construction,
and mean PR survival (~37 min) is roughly the gap between bakes.

Verified safe before trusting it: 8 consecutive bake commits on dev changed
ZERO lines outside newTag/digest/image/artifact-URL. And the four real
admin-app.yaml conflicts measured via merge-tree are each a single `image:`
line — one dev-side digest against four different PR-side digests, no other
content. Newer-wins loses nothing but a stale tag.

The rule is narrow on purpose: a conflict qualifies only if EVERY conflicting
line is a pin marker AND the file is a manifest a bake writes. A hand-edited
replica count in the same file is a real conflict and still ejects
(a_hand_edit_in_the_same_manifest_still_ejects).

Measured over the same 80-PR sample: coverage of conflicting PRs goes from
11% (codegen alone) to 51%. What still ejects is genuine work — lint.rs,
test files, docs — which is the correct verdict.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
triform-admin force-pushed agent/lane-generated-tree-resolver from 2c648ce0f6
All checks were successful
ci / rust-analyzer latency harness (S1 / AC#2) (pull_request) Has been skipped
Details
ci / fmt (pull_request) Successful in 1m17s
Details
ci / build (pull_request) Successful in 2m57s
Details
ci / test (pull_request) Successful in 3m39s
Details
ci / clippy (pull_request) Successful in 3m49s
Details
ci / lane policy mutation proof (pull_request) Successful in 7m20s
Details
to 275c5c32e0
All checks were successful
ci / build (pull_request) Successful in 1m42s
Details
ci / test (pull_request) Successful in 1m24s
Details
ci / fmt (pull_request) Successful in 52s
Details
ci / clippy (pull_request) Successful in 1m18s
Details
ci / rust-analyzer latency harness (S1 / AC#2) (pull_request) Has been skipped
Details
ci / lane policy mutation proof (pull_request) Successful in 4m11s
Details
2026-08-21 08:26:21 +00:00 Compare
triform-admin force-pushed agent/lane-generated-tree-resolver from 275c5c32e0
All checks were successful
ci / build (pull_request) Successful in 1m42s
Details
ci / test (pull_request) Successful in 1m24s
Details
ci / fmt (pull_request) Successful in 52s
Details
ci / clippy (pull_request) Successful in 1m18s
Details
ci / rust-analyzer latency harness (S1 / AC#2) (pull_request) Has been skipped
Details
ci / lane policy mutation proof (pull_request) Successful in 4m11s
Details
to a4683b1876
All checks were successful
ci / build (pull_request) Successful in 1m53s
Details
ci / test (pull_request) Successful in 1m44s
Details
ci / fmt (pull_request) Successful in 38s
Details
ci / clippy (pull_request) Successful in 51s
Details
ci / rust-analyzer latency harness (S1 / AC#2) (pull_request) Has been skipped
Details
ci / lane policy mutation proof (pull_request) Successful in 3m18s
Details
2026-08-23 15:57:09 +00:00 Compare
triform-admin force-pushed agent/lane-generated-tree-resolver from a4683b1876
All checks were successful
ci / build (pull_request) Successful in 1m53s
Details
ci / test (pull_request) Successful in 1m44s
Details
ci / fmt (pull_request) Successful in 38s
Details
ci / clippy (pull_request) Successful in 51s
Details
ci / rust-analyzer latency harness (S1 / AC#2) (pull_request) Has been skipped
Details
ci / lane policy mutation proof (pull_request) Successful in 3m18s
Details
to 68dfc79fad
All checks were successful
ci / build (pull_request) Successful in 57s
Details
ci / test (pull_request) Successful in 1m5s
Details
ci / lane policy mutation proof (pull_request) Successful in 3m13s
Details
ci / fmt (pull_request) Successful in 46s
Details
ci / clippy (pull_request) Successful in 45s
Details
ci / rust-analyzer latency harness (S1 / AC#2) (pull_request) Has been skipped
Details
2026-08-24 15:10:19 +00:00 Compare
All checks were successful
ci / build (pull_request) Successful in 57s
Details
ci / test (pull_request) Successful in 1m5s
Details
ci / lane policy mutation proof (pull_request) Successful in 3m13s
Details
ci / fmt (pull_request) Successful in 46s
Details
ci / clippy (pull_request) Successful in 45s
Details
ci / rust-analyzer latency harness (S1 / AC#2) (pull_request) Has been skipped
Details
This pull request can be merged automatically.
This branch is out-of-date with the base branch
You are not authorized to merge this pull request.
View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u origin agent/lane-generated-tree-resolver:agent/lane-generated-tree-resolver
git switch agent/lane-generated-tree-resolver

Merge

Merge the changes and update on Forgejo.

Warning: The "Autodetect manual merge" setting is not enabled for this repository, you will have to mark this pull request as manually merged afterwards.

git switch main
git merge --no-ff agent/lane-generated-tree-resolver
git switch agent/lane-generated-tree-resolver
git rebase main
git switch main
git merge --ff-only agent/lane-generated-tree-resolver
git switch agent/lane-generated-tree-resolver
git rebase main
git switch main
git merge --no-ff agent/lane-generated-tree-resolver
git switch main
git merge --squash agent/lane-generated-tree-resolver
git switch main
git merge --ff-only agent/lane-generated-tree-resolver
git switch main
git merge agent/lane-generated-tree-resolver
git push origin main
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
triform/cargoless!165
No description provided.